JP4384352B2 - A neuro-fuzzy-based device for programmed hearing aids. - Google Patents

Abstract

A neurofuzzy device (12) is described that provides a fuzzy logic based user-machine interface (9) for optimal fitting of programmable hearing prosthesis (13) using a neural network (4) that generates targets to be matched by the hearing prosthesis (13) based on individual audiometric (1) and other relevant data (2) to the specific impairment and on the neural network (3) accumulated learning from previous successful fittings (8). The incorporated learning process (3) can occur on (16) or off line (14) and implements fitting rationales that can satisfy the needs of a general or specific clientele (8). The parameters of the programmable prosthetic device (6) are set as a group in order to achieve optimal matching to the targets. The user-machine interface realized by a fuzzy logic system (9) deciphers the commands/responses of the user (11) while listening to various stimuli and modifies the targets (5) accordingly thus, providing a closed loop system for in-situ interactive fitting.

Description

[0001]
This application is a continuation-in-part of the provisional application 60/062354 “A Neurofuzzy Methology For Intelligent Hearing Prosthetics” dated October 15, 1997, which gives priority to Insist.
[0002]
FIELD OF THE INVENTION
The present invention is based on information learned from previous successful adaptations and based on user comments / responses while listening in various acoustic environments, towards the adaptation of programmed hearing aids by providing electroacoustic targets based on hearing impairments. It is done. These goals should be compatible with the hearing aid response.
[0003]
BACKGROUND OF THE INVENTION
Programmable hearing aids have a wide range of electroacoustic responses that can accept many different types of hearing impairments, thus providing the possibility of choice for the values of many parameters. An audiologist / hearing aid specialist using audiometry determines the objectives / targets that the hearing device must adapt to compensate for the impairment.
[0004]
With the advent of the programmable listening device, it has become possible to obtain a nearly optimal match between the electroacoustic response and the target. The goal is initially derived by a conventional procedure based on theoretical or empirical considerations proposed for linear, non-programmed listening devices. Recently, new command formulas intended for DSLi / o and non-linear programmed listening devices such as FIG. 6 have been used. Nevertheless, the goal is based on an artificial listening environment (ie noise weighted speech) and does not adequately characterize hearing aid performance in the real environment.
[0005]
An audiologist often faces user complaints that reflect the performance of the device in everyday environments and must adjust the operating parameters of the device with less than optimal instruments and methods. Adjusting parameters one at a time (eg, AGC for low channels, gain for high channels) is not a very optimal procedure because of the large interdependencies between parameters. In addition, user responses and complaints are inaccurate and uncertain and require interpretation by an audiologist.
[0006]
Finally, even instruments that take into account automatic mechanisms for interpreting user responses, such as the fuzzy logic device described in US Pat. No. 5,606,620, are not optimal because they directly manage the individual parameters of the device. Because of this direct management, many conflicting requirements for parameter settings are not adequately resolved by the inherent capabilities of fuzzy logic. Furthermore, the system does not provide an effective mechanism that incorporates learning from successful matches other than manual entry of fuzzy rules.
[0007]
Summary of the Invention
The present invention provides a neurofuzzy device as the first step of the adaptation process, creating an initial goal to compensate for specific hearing loss. These goals are based on both individual acoustic and other data collected and accumulated learning from already successful matches. Starting from a fitting process with effective initial goals, the process can be shortened considerably.
[0008]
A target can have the form of a gain curve, a signal-to-noise curve, etc. for different input levels. These curves are not directly dependent on the particular hearing device used, which represents the “ideal” prosthetic electroacoustic response to a particular disorder.
[0009]
A target is created by a multilayer neural network. This is a “black box” trained to make an optimal fit between a set of acoustic measurements such as hearing thresholds and corresponding best frequencies, and gain curves at different input levels for each subjectivity. It is an information processing system. Neural networks require prioritized knowledge, which requires a large amount of data to converge on the solution.
[0010]
The required priority knowledge is entered into the network during an off-line training phase that takes place during manufacturing, using data collected from hearing aid retailers from selected regions. The currently available normative principles such as NAL and FIG. 6 are used to adapt the starting frequency / gain target curve of a traditional non-programmed hearing aid and programmed device based on such data.
[0011]
It is clear that this known principle is limited in scope because it is impossible to develop a formula based on the observation of large amounts of fit data that sufficiently reflects the interdependencies of the data. Neural networks more effectively acquire the intrinsic nonlinearity of the problem. The acquired knowledge is in the form of a nordal weight in a hidden layer of the network.
[0012]
Network training is an ongoing process that can be done after fine-tuning to achieve a successful fit. A modified target derived from fuzzy logic based on a fine-tuning process described later is used for neural network retraining. Such online training allows a neurofuzzy adaptation device to be biased towards the particularity of a particular client.
[0013]
This neuro-fuzzy method fine-tuning process closes the loop of the fitting process. After the initial goal is created, the settings of all parameters of the hearing aid are guided and communicated to the hearing aid. The user then listens to different acoustic stimuli (eg, different levels of speech, speech and noise at different signal-to-noise ratios, etc.) such as volume, tonality, comfort, distortion, clarity, etc. It is required to exceed the performance of the hearing aid using the quality of acoustic identification. The fuzzy interface changes the target obtained as input of the user response, as well as some target characteristics of the acoustic stimulus (eg total sound pressure level and signal-to-noise ratio) using pre-input rules.
[0014]
Pre-filled fuzzy rules can be provided locally by the manufacturer or by an audiologist. A new modified target is used to derive a new set of values for all parameters, which are downloaded to the hearing aid on the one hand. This cycle is repeated until a new test set is completed and satisfactory results are obtained.
[0015]
Hearing aid parameters are a function of the acoustic properties of the input to the hearing aid and the target curve (which in the extreme is the same as the electroacoustic properties of the hearing aid). This relationship can be encoded into this by pretraining the neural network with the target and input acoustic characteristics and the corresponding output characteristics. Depending on the satisfaction measure at the end of the fitting process, the audiologist / hearing aid specialist can use the final goal for retraining the neural network to create the initial goal.
[0016]
Many other advantages and features of the present invention will become readily apparent from the following detailed description of the invention and its embodiments, from the claims and from the accompanying drawings.
[0017]
Detailed Description of the Preferred Embodiment
While the invention is capable of many different types of embodiments, the disclosure is intended to be illustrative of the principles of the invention and is not intended to limit the invention to the particular embodiments described. With understanding, specific embodiments thereof will now be illustrated and described in detail.
[0018]
The apparatus shown in FIG. 1 is a block diagram representation of the neurofuzzy system of the present invention used to optimize the performance of a programmed hearing aid. This method includes keyboards, touch screens, and computer peripherals such as ports and monitor points used during data entry and acoustic cards and CDROMs used to create acoustic signals used in the fine-tuning process. It can be embodied in a personal computer.
[0019]
The blocks in FIG. 1 are individual subroutines in a software package provided to the audiologist by the hearing aid manufacturer. Hearing data 1 is input to the system via a computer keyboard, which includes hearing threshold data, especially at various frequencies. The above data is input to a pretrained neural network. A portion of this network is shown in FIG.
[0020]
The network creates the required target value required by the hearing aid to adapt to the input level provided to compensate for a particular user's disability. FIG. 2 shows examples of target values for three different frequency ranges. Each of these values is the user's hearing data (only threshold values are shown here for simplicity) and information embedded in the hidden neural layer. This information is placed here during the training period 3 of the neural network. Such training procedures are well known to those skilled in the art.
[0021]
Training of the neural network 3 is performed off-line 14 before the device (software package) is sent to the audiologist for use.
[0022]
If the listening device performance satisfaction level 8 is sufficiently high, the information embedded in the hidden neurons of the neural network is updated and the set of modified targets 15 that are derived after a successful fine-tuning process. It can be used for training. Prior to retraining, the hidden neurons will make an initial set of target curves based on audiometry data 1 and other patient data 2.
[0023]
After retraining, further goal curves will be closer to those obtained from past successful fine tuning processes. Then, for users with similar audiometry data, the system will create an initial goal that is closer to the requirements of this particular impairment, thus shortening the fine tuning process.
[0024]
Each time a new set of targets is created, its parameters must be changed to match the hearing aid's electroacoustic response to that target. Appropriate parameter creation 6 can take the form of a pattern matching search, or it can take the form of a pre-trained neural network in the case of limited consumption.
[0025]
Once the parameters are created, they are sent to the hearing aid 13 via the programmer unit 7. A programmable hearing aid 13 is worn by the user and the user is asked to evaluate its performance while listening to the special audio / acoustic stimulus 12. The user then evaluates quantitatively on several sound quality items such as volume, tonality, comfort, clarity, etc.
[0026]
The user enters the response 11 by entering, for example, a scale number via a touch screen monitor. An example of how this response was repeated is described below. If the stimulus is loud for the user, a large number (3) is entered into the membership graph for the volume scale shown in FIG.
[0027]
A value of 3 on the x-axis of the graph indicates that the sound level identified by the user has a membership level of 1 in the loudness category, and as indicated on the y-axis of the graph. Means membership class 0 in the soft / normal category. The number of membership graphs is input to the fuzzy logic system 9.
[0028]
In FIG. 2, a flow diagram of fuzzy logic and operations for an expert to understand and implement the present invention is shown. The degree of membership in the objective input category called speech level in FIG. 3 indicates that the level of the input signal to the hearing aid is very close to what a normal hearing user calls comfortable (membership degree 0.7). . This same signal is characterized as “loud” by the hearing aid user when indicated by a subjective input called “volume” (membership degree in volume category 0.9). The same signal is characterized as a low frequency on the tonality scale (membership degree 0.8) and on the intelligibility scale as "almost unclear" (membership degree 0.5)
Given the above evaluation, it is clear that the hearing aid in this example does not successfully restore to normal levels of volume and clarity.
[0029]
Some adjustment for initial goals is required. The required adjustments are explained in the following example rule where a fuzzy logic system must be implemented.
1. If the pitch is low or unclear, increase the high frequency.
2. If volume is loud and input is normal, reduce low, medium and high frequencies.
3. If high pitch and clear, reduce high frequency.
4). If the volume is normal and the input is normal, the high, medium and low frequencies are OK.
5. If the pitch is OK and the intelligibility is OK, the high frequency is OK.
6). If the volume is soft and the input is normal, increase the high, medium and low frequencies.
[0030]
The execution of the above example rule is shown in FIG. The result of rule execution will be an increase, decrease or invariance of gain with respect to a frequency / gain curve called normal target that corresponds to the target electroacoustic response of the hearing aid for normal level input sounds. The target curve is divided into low, medium and high frequency parts for simplicity. The gain of each of these parts can be increased, decreased or unchanged depending on the membership value assigned by the above estimation rule.
[0031]
Rule No. 1 assigns two membership degree values (a loudness category on the loudness scale and a comfort category on the speech level scale) to the high frequency increase category in the normal target / high frequency scale. The degree of membership obtained is 0.8. The same procedure is performed for the other rules.
[0032]
If each category is assigned one or more membership degrees, the one with the maximum value is selected. After executing all the rules, it is clear that gains at low and medium frequencies will decrease and gain will hardly change at higher frequencies, based on membership in three additional frequency measures. . The required amount of gain change is derived by a law such as the barycentric law shown in the membership graph for high frequency targets.
[0033]
The shaded part of each department (increase, OK, decrease) gives a visual representation of the degree of membership. The center of gravity of the shaded portion is closer to 3 (close to 1) than −3 on the x-axis. The center of the figure is an intuitively clear value that indicates how many decibels of gain should be added to the high frequency target. Thus, FIG. 3 shows that an unambiguous value of 3 dB should be subtracted from the middle and low frequency gains of the target curve.
[0034]
When a different listening test and evaluation set is completed using the new modified goal, the fine tuning procedure will continue until the audiologist determines that a satisfactory level of performance has been reached. The satisfaction meter 8 can be based on a weighted accumulation change of the target curve. Once the convergence value is obtained, this can be used as an indicator of reaching the diminishing return point of the fine tuning process.
[0035]
Changes and changes can be suggested by those skilled in the art, but all changes and changes that are reasonably and reasonably within the scope of contribution will be guaranteed in order to create and adjust to the initial goal deemed necessary. It is the intention of the present invention to be implemented within the present invention.
[0036]
In view of the above, it will be observed that various changes and modifications can be made within the spirit and scope of the invention. It should be understood that it is not intended or should be estimated to be limited to the particular instrument shown. All modifications within the scope of the claims are, of course, included in the claims.
[Brief description of the drawings]
FIG. 1 is a block diagram of an apparatus for performing a neurofuzzy procedure and points of interaction (data input blocks) with a hearing impaired user and an audiologist.
FIG. 2 is a diagram of a portion of a neural network that creates gain target values for several frequencies based on a listening threshold.
FIG. 3 is a “flow diagram” representation of a fuzzy logic system that changes a goal initially created by a neural network. Fuzzy logic is part of the mutual fine tuning procedure of the listening device.

Claims (2)

A fitting system for programming a separate hearing aid, the fitting system comprising:
A neural network of user data selected to establish initial settings of parameters for the programmable hearing aid, with the processor coupled to circuitry for sending parameters from the processor to the programmable hearing aid and identifying its performance A programmable processor to perform the processing;
Circuitry operable to receive real-time feedback provided by a hearing aid user regarding prerecorded acoustic stimuli and software operable by a processor to provide the prerecorded acoustic stimuli to the hearing aid;
Second software operable by a processor to perform fuzzy logic processing, in response to user multi-parameter feedback, and modifying the hearing aid parameters accordingly;
Said fitting system comprising additional software for downloading modified parameters to a hearing aid for changing performance   The fitting system of claim 1, further comprising software that repeatedly provides acoustic stimuli and repeatedly modifies the parameters in response to user feedback to provide an optimal set of parameters.

Images